LiDAR. Norbert Pfeifer. Institute of Photogrammetry and Remote Sensing (I.P.F.) Vienna University of Technology, Austria

Size: px

Start display at page:

Download "LiDAR. Norbert Pfeifer. Institute of Photogrammetry and Remote Sensing (I.P.F.) Vienna University of Technology, Austria"

Cora Shepherd
6 years ago
Views:

1 LiDAR Norbert Pfeifer Institute of Photogrammetry and Remote Sensing (I.P.F.) Vienna University of Technology, Austria

2 Terms involving Laser Scanning (LS) LASER = Light Amplification by Stimulated Emission of Radiation LIDAR = Light Detection and Ranging (LADAR = Laser Detection and Ranging ) Static Laser Ranging = static LIDAR Kinematic Laser Ranging = Laser Profiling = along track LIDAR Static Laser Scanning / Scanning LIDAR Kinematic Laser Scanning / Scanning LIDAR = along and across track LIDAR LS on different platforms Static mode Terrestrial Laser Scanning (TLS) Kinematic mode Airborne Laser Scanning (ALS) Satellite Laser Ranging (SLR) Mobile Laser Scanning (ground based) 2

3 What this lecture is (not) about This lecture is about LiDAR and LRF. This lecture is not about Laser Scanning. In this lecture LiDAR is embedded into the context of laser scanning topographic situations.

4 Physical principles, motivation LRF: Laser Range Finding Active distance measurement: sensor-object-sensor Possible for none cooperative targets, resp. reflectorless ranging (without prisms) Laser technology is highly developed, therefore there exists the possibility to select an appropriate wavelength Pairing of wavelength and detector (green/red-nir) Strongly focused beam (compare with radar) In more general terms Study of an object by emitting a certain amount of laser energy and by the analysis of the backscattered energy (range, amplitude, etc.) LIDAR = Light Identification Detection Analysis and Ranging (R. Measures) 4

Laser principle Laser = Light Amplification by Stimulated Emission of Radiation (Neodymium: Yttrium Aluminum Garnet) http://en.wikipedia.

5 Laser principle Laser = Light Amplification by Stimulated Emission of Radiation (Neodymium: Yttrium Aluminum Garnet) Output is a beam of light (highly monochromatic, coherent and small beam width). Light is diffusely reflected on object some energy returns to sender measure distance to object d = t/2*c (Distance = time of flight (one way) * speed of light) To measure more than one distance pulses instead of one beam (e.g. by Q- switching; i.e. resonator is turned down while pumping) 5

6 Pulsed Lasers time Laser pulses can be generated by various methods (gain-switching, Q- switching, mode-locking) Typical pulse duration in ALS 2 ns 10 ns 60 cm 3 m pulse length Source: Wagner et al, IAPRS XXX,

7 Pulsed Lasers Pulse Repetition Rate (PRR) Pulse Repetition Frequency (PRF) number of emitted pulses per second Typical values in ALS: khz Example: PRR 100 khz Time between two pulses: 1/100k s = 10-5 s = 0.01 ms Distance between two pulses is 3*10 8 * 10-5 = 3000 m; while flying 1500m above the ground only one pulse is in the air. e.g. PRR = 100 khz, pulse width = 10ns ( 3m) 0.01 ms 10 ns time 7

wikipedia.org/wiki/file:atmospheric_electromagnetic_opacity.

8 Choice of wavelength (λ) λ dependent on laser medium and atmospheric windows: 8 μm 14 μm 3 μm 5 μm (mid IR) 0.7 μm 2.5 μm (near IR), e.g. 1.5 µm, 1064 nm nm 700 nm (visible), e.g. 690 nm, 532 nm ultraviolet otherwise: absorption by water vapor, CO 2, O 3, 8

9 Laser Wavelengths Reflectivity (in %) of water, dry soil, and vegetation Wavelength in use adapted to observed objects (e.g. vegetation and soil) No returns from water different wave length required for bathymetry (usually green lasers) 9

Reflection coefficient for different materials values for 900nm, averaged over typical incidence angles Reflection coefficitent ρ depends in general on material reflectivity and angle of incident

10 Reflection coefficient for different materials values for 900nm, averaged over typical incidence angles Reflection coefficitent ρ depends in general on material reflectivity and angle of incident radiation and view point White paper up to 100% Dimension lumber (pine, clean, dry) 94% Snow 80-90% Beer foam 88% White masonry 85% Limestone, clay up to 75% Newspaper with print 69% Tissue paper, two ply 60% Deciduous trees typ. 60% Coniferous trees typ. 30% Carbonate sand (dry) 57% Carbonate sand (wet) 41% Beach sands, bare areas in dessert typ. 50% Rough wood pallet (clean) 25% Concrete, smooth 24% Asphalt with pebbles 17% Lava 8% Black neoprene 5% Black rubber tire wall 2% Source: Riegl 10

$Laser beam width Beam widens due to diffraction and optical elements Beam waist (smallest diameter) w 0 Definition of beam width for beams with decreasing energy distribution from the center outwards$

11 Laser beam width Beam widens due to diffraction and optical elements Beam waist (smallest diameter) w 0 Definition of beam width for beams with decreasing energy distribution from the center outwards (Gaussian beams) Energy drop to 1/e 2 Alternatively: Energy drop to 1/e, Beam radius expansion w(z) = w 0 (1+(λz/(πw 02 )) 2 ) ½ Divergence β (full angle, beam width 1/e 2 ) is β = 4/π λ/d D = 2w 0 Aperture [Jelalian 1992; Young 2000] typical values: < 1 mrad D 1 beam width β 1/e 2 lateral energy distribution r w(z) z Energy Link: near field far field 11

Energy distribution Energy distribution across (in

, 2003, wavelength 1543nm) Energy distribution along

$, 2004) Beam divergence Diffraction, β λ/d Energy$

12 Energy distribution Energy distribution across (in area) Approximately Gaussian (Jutzi et al., 2003, wavelength 1543nm) Energy distribution along (in time) Approximately Gaussian Riegl -Puls, wavelength 1.5µm (Wagner et al., 2004) Beam divergence Diffraction, β λ/d Energy distributed in time and in area but concentrated in center. 12

Laser equation P P 4πR 4π π β 1 4πR 4π Ω πd 4 2 T R = A ρ η 2 2

T [W] Equally distributed along a sphere [m -2 ] Antenna gain

Backscatter in cone with opening angle Ω Receiver aperture D [m

13 Laser equation P P 4πR 4π π β 1 4πR 4π Ω πd 4 2 T R = A ρ η ATM ηsys + 4 P BK Received power P R [W] Transmitted power P T [W] Equally distributed along a sphere [m -2 ] Antenna gain (opening angle in respect to the sphere) β [],db Area A of target [m 2 ] and with reflection coef. ρ Equally distributed backscatter along a sphere [m -2 ] Backscatter in cone with opening angle Ω Receiver aperture D [m 2 ] Atmospheric and system loss η ATM, η SYS [] Background radiation (sun, shot noise, ) P BK [W] 13

14 Influence of target size on received power P P D 4πβ R 4πAρ Ω 2 R = T 2 4 ηatmη SYS + P BK Influence of target size on received power Extended target A=R 2 βπ/4 P R 1/R 2 Example: open terrain Linear target A=Rβd P R 1/R 3 Example: wire with diameter d (very small) Point target A=const P R 1/R 4 Example: leaf 14

relevant object parameters Isotrop Ω=4π σ = ρa Lambertian Ω=π

15 P Backscatter cross section P D 4πAρ 4πβ R 12 Ω3 σ 2 T R = η 2 4 ATMη SYS + Backscatter cross section σ [m²]: combines all relevant object parameters Isotrop Ω=4π σ = ρa Lambertian Ω=π σ = 4ρA (orthogonal incidence) General: π σ = 4 ρ A Ω P BK A 15

16 LIDAR - LIght Detection And Ranging Pulse Run Time typically used in ALS Pulse-Generator start Counter stop laser Transmitter Δt Detector Photodiode, Photomultiplier, Specular reflection (free) atmosphere Absorption Target Reflector 16

17 Laser Range Finding Range r = c * Δt / 2 with speed of light c ~ 3*10 8 m/s Necessary for successful measurement of Δt Diffuse reflection High SNR (signal to noise ration) Accuracy of time measurement σ t leads to accuracy of range σ r = 1.5*10 8 *σ t, provided c is correct temperature change by 1 C 1 ppm = 1 mm/km Example: σ t = 0.2 ns, σ r = 3 cm Range is measured integral over the entire footprint 17

18 Multiple targets along the laser beam I Laser pulse has a certain energy distribution along (pulse length) and across (foot print area) the beam direction Within the beam: instantaneous field of view multiple different (area, linear, point) targets can be illuminated and can generate a sensable echo One pulse can therefore generate multiple echoes i.e.: first echo, intermediate echoes, last echo 18

19 Cross section and height spread Slanted and rough surfaces = Echoes are widened Two echoes Cross section is a function of range σ σ(r) 19

20 Discrete return systems In discrete return laser scanning systems, ranges to objects separated further than half the echo length can be measured Objects closer together cause a compound echo, and the range will refer to an intermediate distance According to their arrival time, echoes are called first, second, intermediate, last echo Scanning over a forest, last echoes may be Reflections from the ground Reflections from dense vegetation above the ground long ranges, referring to ranges beyond the ground, e.g. due to multi-path effects 20

21 First and last echoes DSM of first echoes DSM of last echoes Leitha: GE0A0209.xyz 21

22 Discrete Echo vs. Full-Wave-Form LIDAR Echo waveform Discrete echoes Laser foot print: ~ 0.2 3m Recording the entire return signal using a sampling interval of ~1ns FWF ALS 22

23 Discrete Echo vs. Full-Wave-Form LIDAR 23

24 Backscattering cross section How is the echo formed? Lidar equation Power received, P R (t) vs. transmitted power P T (t) P P D 4πβ R 4πAρ Ω 2 R = T 2 4 ηatmη SYS + FWF realizes power as a function of time (~range) Height spread of object should be considered, too Therefore, target cross section is a function of range (~time), too σ σ(r) P BK

25 After: Wagner. Der Laserstrahl und seine Interation mit Oberflächen unterschiedlicher Beschaffenheit. Inst. of Photogrammetry and Remote Sensing, Vienna University of Technology, Pulse deformation on objects Flat surface Sloped surface Tree Emitted pulse Height spread of object Re-ceived echo

26 Target cross section Received echo is a convolution of the emitted pulse with the target cross section Multiple, spatially distinct echoes, may be separated in the Lidar equation For one cross section σ i (R), the Lidar equation is: P 2 Ri +δ D 2R R, i ( t) PT ( t ) σ 2 4 i 4πβ R c i R δ ( R) dr The entire echo is the sum over the individual returns i 2δ is the spatial extent along the beam

27 Terms P 2 Ri +δ D 2R R, i ( t) PT ( t ) σ 2 4 i 4πβ R c i R δ Dynamic Lidar equation σ i (R) is the differential target cross section σ i (R) dr, the integral over the diff.c.s., is the target cross section i ( R) dr Target cross section = backscattering cross section = cross section

Convolution real example Viewing direction Emitted pulse Height

06 0.04 Cross section [m 2 ] Amplitude 25 20 15 10 Laser pulse

02 5 0 534 535 536 537 538 539 540 541 542 Distance (m) 0 534 535

28 Convolution real example Viewing direction Emitted pulse Height distribution of objects Received echoes Amplitude Cross section [m 2 ] Amplitude Laser pulse Gaussian model Distance (m) Distance (m) Convolution Viewing direction Viewing direction Data courtesy of Riegl Gmbh.

29 Full waveform analysis 2 options Extract parameters of interest directly from the waveform - simpler, but waveform depends on object and mission parameters 2 step approach 1. extract cross section (and/or differential cross section) parameters 2. extract parameters of interest from cross section parameters mathematically more complex, but cross section independent of mission parameters

30 Waveform parameters Waveform parameters include Duration (length), Centroid (mean elevation) Distance from first to last peak Height of median energy (HOME) Return waveform energy The above parameters can be derived from the sampled waveform directly. Used in analysis of ICESat data! Images: Ph.D. Duong, TU Delft

31 Waveform modeling Extracting waveform modes and parameters of each mode (=echo) is better performed by waveform modeling. - Overlapping echoes can be separated - Noise reduction through fitting of functional models

32 Waveform modeling Waveform modeling is performed similarly by different groups [Delft: Van Duong, Vienna: Wagner, Newcastle: Yu-Chin Lin, Paris: Mallet, Colorado: Lefsky, NASA: Harding; et.al.] GMM (Gaussian Mixture Model) is applied most often (or extensions thereof). wf ( t) = c + i= 1 ( t μ The wf (waveform) is a sum of Gaussians plus a constant offset c. m A e i i ) 2 / s 2 i

33 Waveform modeling GMM wf ( t) = c + m i= 1 A e i ( t μ ) i 2 / s 2 i The number of modes (echoes?) has to be estimated in advance (sets model complexity). The estimation is not linear, thus approximate values for the unknowns have to be supplied. The unknowns are: c, A, μ, s ; i = 1, K i i i m

34 Meaning of parameters Waveform modeling - c average background noise (electronic noise) - µ arrival time of backscattered pulse (subtract from µ of emitted pulse to obtain pulse travel duration) range - A power of received echo (amplitude) - s width of received echo (must not be smaller than s of emitted pulse)

Full-Wave-Form-Analysis in Postprocessing Gaussian decomposition: Detection of echos by fitting Gaussian curves Information per echo: Amplitude P Range R Echo width s R s P Cross section σ With

35 Full-Wave-Form-Analysis in Postprocessing Gaussian decomposition: Detection of echos by fitting Gaussian curves Information per echo: Amplitude P Range R Echo width s R s P Cross section σ With FWF-Analysis the echos from the lower vegetation can be separated from the echos from the ground much better or the mixted echo of both objects is detected because of its larger echo width. Improvement of DTM computation 35

36 FWF Example Schönbrunn Example on next slides shows one ALS FWF strip over Schönbrunn castle (Vienna, Austria) acquired with a Riegl laser scanner. Orthophoto 36

37 FWF Data Range Amplitude Echo width Ortho photo 37

38 Notes Full waveform analysis Emitted Gaussian pulse (system waveform) shape can be engineered by system designer Small deviations may remain (blue: typical system waveform, red: ideal Gaussian) Assumption on Gaussian differential cross section becomes less and less valid for larger footprints due to increase of object complexity within footprint. deconvolution requires some form of regularization / noise suppression. Mathematically difficult problem with physical side conditions 38

39 Full waveform analysis Deriving the cross section makes the analysis independent of the emitted pulse power, the pulse duration, and the range. Thus it is to be preferred. But deconvolution is complex and does not necessarily have a unique solution. If, however, the emitted pulse is Gaussian and the differential target cross sections are Gaussians, then the backscattered echo follows the GMM. (because Gaussian convolved with Gaussian is Gaussian) P 2 Ri +δ D 2R R, i ( t) PT ( t ) σ 2 4 i 4πβ R c i Ri δ m 2 2 ( t μi ) / si ( t) = c + Ai e i= 1 wf ( R) dr

40 Full waveform analysis Convolution and deconvolution Variance (s 2 ) is square of width measure - s 2 E for emitted pulse - s D2 for detected echo - s CS2 for cross section (width) s E2 + s CS2 = s 2 D Estimated from FWF data: s E2, s 2 D Therefore: s 2 CS = s 2 D -s 2 E Amplitude Laser pulse Gaussian model Distance (m) Amplitude Distance (m)

41 Cross section Cross section (differential cross section) is found? - Yes, w.r.t. cross section width - No, w.r.t. integral under the curve System parameters unknown - Require radiometric control point(s)

42 Radiometric Calibration Laser equation in backscr. cross sect. σ[m²]: Backscatter coefficient [db]: P γ σ 4σ = 2 Acosα = πβ R P D 4πβ R 2 T R = σ η 2 4 ATMη SYS + 2 P BK Laser equation in γ: P P D 16R 2 R = T 2 γ ηatmη SYS + P BK Beam cross section: Acosα Acosα α α A 42

Radiometric Calibration Uncalibrated radiometric information difficult to use 2 PD t = ηatmη sys 16R Pr 2 γ from FWF analysis: P t P r Ss ˆ s Ps ˆ p γ = D 2 16 Sˆ s s η sys R 2 η

.. atmospheric transmission factor γ... Backscattering coefficient [m 2 m -2 ] Ŝ... Amplitude of the system waveform [DN] s s... Standard deviation of the system waveform [s] P.

43 Radiometric Calibration Uncalibrated radiometric information difficult to use 2 PD t = ηatmη sys 16R Pr 2 γ from FWF analysis: P t P r Ss ˆ s Ps ˆ p γ = D 2 16 Sˆ s s η sys R 2 η Pˆ atm s p P r... Received power [W] P t... Transmitted power [W] D... Diameter of receiver aperture [m] R... Range [m] η sys... system transmission factor η atm... atmospheric transmission factor γ... Backscattering coefficient [m 2 m -2 ] Ŝ... Amplitude of the system waveform [DN] s s... Standard deviation of the system waveform [s] P... Amplitude of the echo [DN] s p... Standard deviation of the echo [s] C cal (only dependent on constant system parameters) η = 10 2R a /10000 atm a... atmospheric attenuation coefficient [db/km] j 43

44 Resulting radiometric quantities Backscattering coefficient γ []: γ = C cal atm Assuming diffuse (Lambertian) reflectors: R 2 η Pˆ s p γ d = 4ρ d cosα with ρ d the diffuse reflectance measure (0 to 100%), which depends only on the object and not on the view point. If C cal is known, then for each target with known incidence angle α its reflectance measure ρ d can be computed: ~ ρ d = 4 γ = cosα C cal 4 R 2 Pˆ s p cosα η atm 44

Determination of C cal Calibration Target (CT) (e.g. asphalt) γ CT, j = 4 ~ ρ CT ( α ) cosα j j γ = C cal R 2 η Pˆ atm s p γ CT,j... Backscattering coefficient of the CT of echo j [m 2 m -2 ] ρ CT.

45 Determination of C cal Calibration Target (CT) (e.g. asphalt) γ CT, j = 4 ~ ρ CT ( α ) cosα j j γ = C cal R 2 η Pˆ atm s p γ CT,j... Backscattering coefficient of the CT of echo j [m 2 m -2 ] ρ CT... Reflectance of the calibration target CT θ j Angle of incidence of the echo j within the calibration target Idea: Using a reflectometer ρ CT can be determined for the asphalt, which is then used for the ALS measurements to derive C cal. From then on, all γ from ALS can be converted into ρ. 45

46 Determination of C cal for 1 Calibration Target (CT) Fieldin strip1 Fieldin strip2 Fieldin strip3 P γ Vienna wide ALS campaign December 2006 ρ d 46

47 Amplitude P (Vienna 2006) 47

48 Diffuse reflectance ρ (Vienna 2006) 48

49 Is there a use in full waveform?

50 ALS Leithagebirge DSM (first echo) In cooperation with: Dr. Michael Doneus, Inst. f. Ur- u. Frühgeschichte, Wien 50

51 ALS Leithagebirge DSM (last echo - SCOP++) In cooperation with: Dr. Michael Doneus, Inst. f. Ur- u. Frühgeschichte, Wien 51

52 ALS Leithagebirge Aerial images In cooperation with: Dr. Michael Doneus, Inst. f. Ur- u. Frühgeschichte, Wien 52

53 ALS Leithagebirge FWF Attribut: Amplitude First-Echo In cooperation with: Dr. Michael Doneus, Inst. f. Ur- u. Frühgeschichte, Wien 53

54 ALS Leithagebirge FWF Attribut: Echo Width In cooperation with: Dr. Michael Doneus, Inst. f. Ur- u. Frühgeschichte, Wien Last-Echo 54

55 Improved DTM derivation using ALS point clouds ALS point cloud (combiend from all ALS flight strips) Selection of the Last Echo points Pre-Filtering / Pre-Classification using the FWF attributes (elimination of potential vegetation echos) Filtering / Classification (Seperation of terrain and off-terrain points; e.g. using robust filtering) Computation of the DTM using all points classified as terrain 55

function Accuracy of extracted echo width depends on amplitude (the higher the

56 Hard classification vs. smart priors Scatter plot of last echoes from ground under vegetation Weight function Accuracy of extracted echo width depends on amplitude (the higher the better). Prior weights depend. on echo width and ampl. instead of hard classification. Echo width of transmitted signal Echo width [ns] Mücke W.,

57 Robust Interpolation without individual weights Kriging profile (black) without robust interpolation, equal weighted points Kriging profile (black) after rob. interpolation (10 iterations) 57

58 Robust Interpolation with individual weights Kriging profile (black) without robust interpolation, individual weights per point (echo width) Kriging profile (black) after rob. interpolation (2 iterations) 58

59 Improved DTM after considering the Echo Width DONEUS M., BRIESE C. Full-waveform airborne laser scanning as a tool for archaeological reconnaissance, International Conference on Remote Sensing in Archaeology, Rom; ; in: "From Space to Place. Proceedings of the 2nd International Conference on Remote Sensing in Archaeology", BAR International Series, 1568 (2006), , December

60 Summary Fundamentals of LiDAR for geo-spatial applications were presented Next step: from physics to geometry Still missing in this introduction - What to do, if the signals are not approximately Gaussians? - How does this apply to phase-based laser ranging? - Direct geo-referencing - Scanning as such - Strip adjustment

CALIBRATION OF A TERRESTRIAL FULL WAVEFORM LASER SCANNER INTRODUCTION

CALIBRATION OF A TERRESTRIAL FULL WAVEFORM LASER SCANNER Preston J. Hartzell Craig L. Glennie Department of Civil and Environmental Engineering University of Houston Houston, TX 77204 pjhartzell@uh.edu